Blair Witch

Workflow

1. 01

   Scene analysis

   The film’s own scenes came first: their light, tone and rhythm were studied before anything was generated.

2. 02

   Image development

   Key frames for the figure, the forest and the transformations were generated and judged against the live-action footage.

3. 03

   Animation and camera

   Approved frames were animated, with camera and environmental movement designed for each scene.

4. 04

   Edit, visual effects and sound

   The shots were cut, colour-matched and synced to sound so that they sit inside the film.

What was human, what was generated

Blair Witch is a live-action film with generative AI inside it. People wrote, shot and edited the film. Güneş Polater directed and produced the AI sequences: he decided what each one shows and matched it to the footage around it. AI generated the pictures inside those sequences; the edit, the visual effects work, the colour match and the sound sync were done by hand at Leopati in Ankara, Türkiye.

Prompt approach

Each shot began from the film, not from a blank prompt. The scene it had to join set the light, the lens and the weather; the generated element was described only as what changes inside that frame. Every result was judged against the live-action footage on either side of it.

Problems and solutions

1. Problem A generated shot cut next to live action shows its seam: different light, different texture, different motion.

   Solution Each sequence was built from the scene it joins and checked against it shot by shot; colour and sound were matched in post-production.

2. Problem A transformation, food turning to flesh, can look like an effect instead of an event.

   Solution The table shot holds one framing from start to finish, so the change happens in front of the audience with no cut to hide behind.